Risk Analysis DOI: 10.1111/risa.12540

Survey of Ambient Air Pollution Health Risk Assessment Tools

Susan C. Anenberg,1,∗ Anna Belova,2 Jørgen Brandt,3 Neal Fann,4 Sue Greco,5 Sarath Guttikunda,6 Marie-Eve Heroux,7 Fintan Hurley,8 Michal Krzyzanowski,9 Sylvia Medina,10 Brian Miller,8 Kiran Pandey,11 Joachim Roos,12 and Rita Van Dingenen13

Designing air quality policies that improve public health can benefit from information about air pollution health risks and impacts, which include respiratory and cardiovascular diseases and premature death. Several computer-based tools help automate air pollution health im- pact assessments and are being used for a variety of contexts. Expanding information gath- ered for a May 2014 World Health Organization expert meeting, we survey 12 multinational air pollution health impact assessment tools, categorize them according to key technical and operational characteristics, and identify limitations and challenges. Key characteristics in- clude spatial resolution, pollutants and health effect outcomes evaluated, and method for characterizing population exposure, as well as tool format, accessibility, complexity, and de- gree of peer review and application in policy contexts. While many of the tools use com- mon data sources for concentration-response associations, population, and baseline mortality rates, they vary in the exposure information source, format, and degree of technical complex- ity. We find that there is an important tradeoff between technical refinement and accessibility for a broad range of applications. Analysts should apply tools that provide the appropriate geographic scope, resolution, and maximum degree of technical rigor for the intended assess- ment, within resources constraints. A systematic intercomparison of the tools’ inputs, assump- tions, calculations, and results would be helpful to determine the appropriateness of each for different types of assessment. Future work would benefit from accounting for multiple un- certainty sources and integrating ambient air pollution health impact assessment tools with those addressing other related health risks (e.g., smoking, indoor pollution, climate change, vehicle accidents, physical activity).

KEY WORDS: Air pollution; health impact assessment tools; mortality; ozone; particulate matter

1Environmental Health Analytics, LLC, Washington, DC, USA. 9Environmental Research Group, King’s College London, Lon- 2Abt Associates, Bethesda, MD, USA. don, UK. 3Department of Environmental Science, Aarhus University, 10French Institute for Public Health Surveillance, Saint Maurice, Roskilde, Denmark. France. 4U.S. Environmental Protection Agency, Research Triangle Park, 11The World Bank, Washington, DC, USA. NC, USA. 12Institute of Energy Economics and Rational Use of Energy, Uni- 5Public Health Ontario, Toronto, Ontario, Canada. versity Stuttgart, Stuttgart, Germany. 6Division of Atmospheric Sciences, Desert Research Institute, 13European Commission, Joint Research Centre (JRC), Institute Reno, NV, USA. for Environment and Sustainability (IES), Ispra VA, Italy. 7World Health Organization Regional Office for Europe, Bonn, ∗Address correspondence to Susan Anenberg, Environmental Germany. Health Analytics, LLC, 3704 Ingomar St. NW, Washington, DC, 8Institute of Occupational Medicine, Edinburgh, UK. USA; [email protected].

1 0272-4332/16/0100-0001$22.00/1 C 2016 Society for Risk Analysis 2 Anenberg et al.

1. INTRODUCTION concentration-response relationships from epidemi- ology studies in one location among a single popula- Decades of toxicological, clinical, and epidemi- tion at low concentrations only. The IER approach ological research demonstrate significant associa- was used by the Global Burden of Disease project to tions between exposure to ambient air pollution and calculate that approximately 3.2 million and 150,000 deleterious human health effects, including respira- premature deaths globally in 2010 were attributable tory disease, cardiovascular disease, and premature to ambient PM and ozone, respectively.(16) In addi- (1–4) 2.5 death. Air pollution is a mixture of components, tion, household air pollution was associated with an including fine particles (PM2.5), ground-level ozone estimated 4 million premature deaths globally in 2010 (O3), oxides of nitrogen (NOx), and oxides of sulfur (approximately 0.5 million overlapping with the esti- (SOx). The health effects of individual air pollutants mated ambient particulate matter deaths), making it have been reviewed in detail to support the setting of the fourth worst health risk factor after high blood ambient air quality guidelines by the World Health pressure, smoking, and alcohol use. Ambient PM , (5) 2.5 Organization and national standards, such as the not including ambient ozone, was the seventh worst (6–9) U.S. National Ambient Air Quality Standards. health risk factor globally based on associated pre- For example, extensive reviews by the U.S. EPA con- mature deaths. clude that long-term exposure to PM2.5 is associated Driven by the extensive body of literature with premature death, heart attacks, irregular heart- demonstrating their health effects, ambient concen- beat, and respiratory symptoms such as aggravated trations of particulate matter and ozone are now (7) asthma and decreased lung function, that short- regulated in many countries. Setting these regu- term exposure to O3 is associated with respiratory ef- lations at levels sufficient to protect public health fects, cardiovascular effects, and premature all-cause typically makes use of a variety of technical inputs, mortality, and that long-term exposure to O3 is as- including estimates of the total population health sociated with respiratory effects, including some ev- burden posed by the pollutants at current concen- idence for association with premature respiratory trations as well as the health benefits of reducing air (8) mortality. Beyond individual pollutants, research is pollution levels. There are a number of variants to also increasingly demonstrating the health effects of these two inputs—for example, analysts may wish to air pollution mixtures, such as from biomass smoke understand the historical trend in the human health (10–12) or traffic. In addition, studies show that some burden of air pollution caused by a specific polluting areas have a confluence of health risks from multiple sector or experienced by a certain subpopulation pollutants, and therefore a multi-pollutant air quality such as children. Over the last decade, governmental, management approach may be efficient at mitigating intergovernmental, and nongovernmental entities (13,14) those risks. have invested in tools that are better able to meet While the majority of epidemiological research this growing demand for more specific and timely has been conducted in North America and Europe, information regarding health impacts associated with insofar as it is available, evidence suggests that the exposure to air pollutants. For example, the U.S. En- associations between air pollutants and health effects vironmental Protection Agency developed the Envi- (10) are relatively consistent around the world. How- ronmental Benefits Mapping and Analysis Program ever, as some studies indicate a leveling off of risk (BenMAP-CE) in part to help fulfill requirements by at high PM2.5 concentrations found in many devel- the Office of Management and Budget and the Clean oping countries, researchers have leveraged epidemi- Air Act to characterize the benefits and costs of U.S. ological studies of ambient PM2.5, typically higher air pollution regulations.(17,18) Other countries and levels of indoor air pollution in developing coun- intergovernmental organizations such as the World tries where solid fuel is used for cooking and home Health Organization and World Bank have invested heating, and very high particulate exposure levels in similar tools to quantify air-pollution-related from cigarette smoking to develop “integrated ex- health impacts for a variety of purposes.(19–21) (15) posure response” (IER) curves. Because the IER Health impact and health burden assessments curves were developed from concentration-response depend strongly on the evidence available from air data taken from around the world, among a vari- pollution epidemiology and exposure science. Re- ety of populations, and across a range of exposure cent advances in these two disciplines have enabled concentrations including high PM2.5 concentrations, health impact assessments to combine findings from they may be more broadly applicable globally than atmospheric science and epidemiology, allowing Survey of Ambient Air Pollution Health Risk Assessment Tools 3 analysts to quantify an increasing number of health ing health impacts, wherein epidemiology-derived outcomes in far greater detail than was previously concentration-response associations and population- possible. Over the last decades, air pollution epi- level exposure estimates are used to determine the demiology has characterized risks of certain health portion of cases of a particular health effect that may outcomes in a population exposed to a higher level of be attributable to air pollution in a particular time pe- air pollution relative to a population exposed to less riod. This method requires information about air pol- air pollution (“relative risk”). Many determinants lution concentration levels, the relationship between of health, including socioeconomic status and other concentrations and health outcomes (which could be risk factors such as smoking, affect the same health provided by systematic analysis of studies done out- outcomes as does outdoor air pollution, and may co- side of the assessed population,(47) or by a shrunken vary with pollution. Epidemiology studies attempt estimate analysis combining a robust meta-analytical to isolate the effect of air pollution by controlling risk with a local one when available(48)), and the char- for such confounders, either by design (as in studies acteristics of the populations exposed, including their of short-term exposure) or in the analysis (as in baseline health status, age, and location (Fig. 1). studies of long-term exposure and mortality). Thus, Air pollution health risk assessment tools are the resulting relative risk estimates for air pollution typically preloaded with health and demographic may be considered to be largely independent of data and concentration-response associations, and all the confounders included in the study’s model some allow for user-specified inputs. Some of these (other unknown confounders may not have been tools also have built-in air pollution exposure infor- included and could thus still influence the results). mation connecting emissions to the exposure metric, Building from this base of air pollution relative risk requiring users to input only information about emis- estimates, quantitative air pollution health impact sion changes; others read in user-specified exposure assessments can now be performed at various scales estimates. To determine the appropriate air pollution and resolutions for many air pollutants, including health impact assessment tool, data sets, and context PM2.5,O3,NOx, and SOx. for interpreting results, analyses typically begin with A variety of studies have now quantified health key demographic and economic data for the relevant impacts associated with these air pollutants at geographic area and population, including per capita global,(16,22–28) regional,(29–32) national,(17,33–37) and income, health-care delivery systems, prevalence of local scales.(13,17,18,38–43) Quantification of changes smoking, climate (including use of air conditioning), in health effects due to various emission reduction use of combustion sources indoors (e.g., for cooking scenarios has been the basis of analysis support- and heating), the nature of the air quality monitoring ing air quality policy development of the European system, and major health indices. The availability of Union(44,45) and the United States,(46) in addition to high-quality data sets for these parameters varies by other countries. Results of these assessments are of- context, including country and spatial scale. ten reported in numbers of deaths and disease cases, This article reviews 12 air pollution health years of life years lost (YLL), disability-adjusted life impact assessment tools that are currently available, years (DALYs), or change in life expectancy at- categorizes the tools according to key technical and tributable to total air pollution concentrations or a operational characteristics for different assessment change in air pollution concentrations. contexts, and identifies information gaps relevant Using computer programs, or tools, to auto- for future work. These tools, often designed for mate the procedure for calculating the incidence a particular type of assessment context, vary in or prevalence of air-pollution-related health impacts methodological approach, technical complexity, geo- offers several advantages: simplicity (lowering the graphical scope, resolution, and other aspects. Here barrier of entry for new analysts to conduct as- we define “assessment context” as the parameters of sessments), consistency, comparability among assess- the intended analysis, including its purpose (e.g., is ments, and quality assurance. However, as discussed it being used to inform the setting of a policy or as a later, there may be associated disadvantages also, nonregulatory communication tool), the geographic especially if the analyst is unaware of the detailed area of interest (e.g., a single city, a country, a region assumptions built into the tool and/or of their im- of the world, or global), and the type of information portance in particular applications. Many of these it seeks to provide (e.g., health burden of current available air pollution health impact assessment tools air pollution levels, health benefits of reduced use the attributable fraction approach to quantify- air pollution levels, health benefits of reduced air 4 Anenberg et al.

Fig. 1. Schematic of air pollution health impact assessment method and typical data inputs. pollution emissions). This article is the first to survey tematically compared the results across many tools available air pollution health risk assessment tools (e.g., by comparing results from various tools using to begin to understand the spectrum of methods a consistent set of analyses), as has been done for air and assumptions used. To date, no study has sys- quality models that are often used as inputs to health Survey of Ambient Air Pollution Health Risk Assessment Tools 5 impact assessment tools.(49–51) In general, air quality stand-alone computer programs that are packaged models have been reported on in the peer-reviewed for distribution and use by analysts other than the de- literature in far greater detail compared with air velopers) in an English-language peer-reviewed jour- pollution health impact assessment tools, allowing nal article. for the community of researchers to understand their The tool developers were then asked to re- inputs, processes, and outputs. Given the relative spond to a list of survey questions on technical paucity of public information about air pollution and operational aspects of their tool (see Supple- health impact assessment tools to date, this first mental Material). The survey questions covered a survey of the available tools will enable comparisons broad range of characteristics, such as the pollu- across the tools to be made in the future. tants included, geographical scope, spatial resolution, Tools that as currently configured apply to a sin- temporal resolution, exposure metric, exposure in- gle country are summarized in the Supplemental Ma- formation source, health outcomes, concentration- terial but are not synthesized in this article due to response associations, population and baseline inci- their limited geographical scope. This article does dence and/or prevalence data sources, format of the not address methods to assign an economic value to tool, whether it is opensource or proprietary, how health outcomes, though many of the tools reviewed to obtain the tool, whether training materials exist, include that capability. In addition, the tools re- whether the tool has been peer reviewed, whether viewed here focus on ambient air pollution, as meth- the tool has been used to support an air quality pol- ods and tools for quantifying household air pollution icy, and the developing institution and contact per- health impacts are in an earlier stage of development. son. Of the 20 tools, information was submitted in response to the survey for 17. The 17 tools for which information was pro- vided were categorized by geographical scope ac- 2. METHODS cording to the tool’s preconfiguration. Geographical This article was first developed as a white pa- scope is often the first factor an analyst must con- per for input to the World Health Organization sider in selecting a tool for a particular assessment. (WHO) Expert Meeting on Health Risk Assessment Geographic scope is defined as the spatial coverage held in Bonn, Germany, May 12–13, 2014, and was or extent of the tool as currently configured, and is subsequently revised and expanded with additional distinct from spatial resolution. For example, a tool information.(52) To identify the universe of relevant with global scope may have a national resolution tools that are currently available and in use, the steer- (i.e., including countries around the world) or city ing committee for the expert meeting was asked to resolution (i.e., including cities around the world). submit the names of air pollution health risk assess- Tools with regional scope are those that include mul- ment tools that their organizations have developed, tiple countries in one or more discrete world regions. used, or know others have used. The steering com- Tools that were available for individual countries are mittee consisted of scientists and policy analysts from not reviewed here due to their limited geographical WHO, academia, governments, and the World Bank. scope. Five national scope tools are summarized in The developers of the identified tools were then con- the Supplemental Material and include the Air Qual- tacted to determine interest in participating in this ity Benefits Assessment Tool (AQBAT) and the Ill- survey. Some of the tool developers submitted ad- ness Cost of Air Pollution tool (ICAP) for Canada, ditional tools for inclusion in the white paper. This the Integrated Transport and Health Impact Model- process yielded 20 tools used to quantify ambient air ing Tool (ITHIM) for the United Kingdom, and the pollution health risks at various geographic scales. Co-Benefits Risk Assessment Screening Model (CO- We then performed an informal review of the peer- BRA) and AP2 (formerly APEEP) model for the reviewed literature published in English using the United States. following search terms together in Google Scholar: “air pollution,” “health impact” or “health risk,” and 3. KEY TECHNICAL CHARACTERISTICS “tool.” This process did not uncover any additional tools currently in use that also met our criterion of Of the 12 tools, nine have global scope, en- encompassing multiple countries, though there may compassing countries and/or cities around the world be additional tools that meet our criteria that have (Tables I–III). Four of these tools are designed not been described as a tool (considered here as to be flexible in scope and can be used for 6 Anenberg et al.

Table I. Key Technical Characteristics of Tools with Global Scope

Characteristic AirCountsTM AIRQ2.2 BenMAP-CE EBD GMAPS IOMLIFET LEAP-IBC SIM-Air TM5-FASST

Spatial resolution: Regional x x x x x x National x x x x x x x City-level x x x x x x x Pollutants: Any grid x x x x a PM2.5 x (primary) x x x x x x x PM10 xxxxx Ozone x x x x x NO2 xx x x SO2 xx x CO x Other Black Any affecting smoke mortality Health outcome: Mortality (cases) x x x x x x x x x Disability-adjusted xxxxx x life years (DALY) or years of life lost (YLL) Morbidity (cases) x x x x x aThe SIM-air framework outputs all the criteria pollutants, with linkages for use of all the relevant pollutants in the regional/urban chemical transport models. Only in case of the health impacts, PM is considered as the target pollutant.

TM analyses ranging from the local to global resolu- (PM2.5) impacts, though AirCounts includes only tions (AirQ2.2, BenMAP-CE, EBD, IOMLIFET). primary PM2.5 (excluding secondarily formed sulfate, The three remaining tools apply to a specific re- nitrate, and secondary organic aerosols). Most tools gion of the world encompassing several countries are readily able to estimate coarse particulate matter (Tables IV–VI). Summary descriptions of the 12 (PM10) and ozone impacts, and some include NOx, tools are included in the Supplemental Material. SOx, and CO. A few tools also include other pollu- After having provisionally selected a tool based tants such as heavy metals and black smoke. on its predefined geographic scope, the analyst Similarly, all the tools reviewed here calculate would also consider: how spatially resolved the impacts of air pollution on premature mortality in impact estimates are (region/nation/administrative terms of the number of excess or avoided deaths. boundary), which pollutants and health effect out- However, most tools are set up to quantify “all comes the tool is preconfigured to assess, the cause mortality,” or deaths from any cause, including method for characterizing population exposure those that are associated with air pollution, such as (Sections 3.1 and 3.2), and choice of concentration- cardiovascular disease, as well as those that are not response functions and demographic characteristics associated with air pollution, such as accidents and (Section 3.3). Additional operational factors may infectious disease. In most cases, particularly for also be important: format, accessibility, complexity, developing countries, it is necessary to extrapolate and degree of peer review and application in policy epidemiologically-derived relative risk estimates contexts (Section 4). The following sections describe from one country to another given the lack of each of these key characteristics. air pollution epidemiology studies in many coun- tries and often greater confidence in estimates of 3.1. Pollutants and Health Effect Outcomes concentration-mortality risk associations based on all available studies as compared to a smaller number The tools reviewed here differ in terms of pollu- of studies in a particular location. In such cases, it tants addressed and health outcomes quantified. All may be more appropriate to quantify cause-specific tools reviewed in this article except one (GMAPS) mortality, such as cardiovascular or respiratory are preconfigured to assess fine particulate matter mortality, rather than all-cause mortality, as the Survey of Ambient Air Pollution Health Risk Assessment Tools 7

Table II. PM2.5 Concentration-Response, Population, and Baseline Incidence Data Sources for Tools with Global Scope

Characteristic AirCountsTM AIRQ2.2 BenMAP-CE EBD GMAPS IOMLIFET LEAP-IBC SIM-Air TM5-FASST

PM2.5 concentration- response relationship: User-defined x x x x American Cancer xxx xxxxx Society(3,56) GBD Integrated xx Inprepx Exposure Response(15) Other x x Population data source: User-defined x x x x United Nations x x CIESINa xxxxx Other x Baseline incidence data source: User-defined x x x World Health xxx xx Organization Other xx aCenter for International Earth Science Information Network (CIESIN), available at www.ciesin.org.

Table III. Key Operational Characteristics of Tools with Global Scope

Characteristic AirCountsTM AIRQ2.2 BenMAP-CE EBD GMAPS IOMLIFET LEAP-IBC SIM-Air TM5-FASST

Format: Software download x x Microsoft office xx x x x x program Web based x In prep In prep Open source x x x x x x x In prep Proprietary x x Peer reviewed/policy applications: Peer reviewed In prep Expert x x In prep x In prep x In prep Used for policy xx xxxxx applications

underlying mix of causes of death can vary greatly YLLs, DALYs, and morbidity cases (e.g., respira- between populations. The recently published IER tory and cardiovascular hospital admissions, cases of curves connecting PM2.5 concentrations to causes chronic obstructive pulmonary disorder). Most tools, of death (ischemic heart disease, cerebrovascular such as BenMAP-CE and the LEAP-Integrated disease, chronic obstructive pulmonary diseases, Benefits Calculator (LEAP-IBC), estimate impacts lung cancer, and acute lower respiratory infections) attributable to air quality changes in a single year, may now be used for many applications where though these impacts may lag over a multi-year population-specific risk information is unavailable, period. The IOMLIFET model can characterize the as they draw from studies around the world and change in the risk of premature death among a cohort across a wide range of concentrations.(14,53) These of individuals over the course of their lifetime.(54) curves have not yet been parameterized in all the Quantifying air-pollution-related morbidity tools described here. Many tools can also quantify impacts around the world is made difficult by the 8 Anenberg et al.

Table IV. Key Technical Characteristics of Tools with Regional an external source (Table VII). Methods for char- (i.e., Multiple Countries in One World Region) Scope acterizing exposure often determine the spatial res- Characteristic Aphekom EVA EcoSense olution at which air-pollution-related health impacts are calculated and results reported. Some tools assign Region: Europe Northern Europe air quality values to a grid, wherein the geographi- Hemisphere cal scope is divided into cells (either uniform or vari- National x x able in shape) and population exposure and health City-level x x x Grid x x impacts are quantified separately for each cell. Other Pollutants: tools assign air quality data to geopolitical bound- PM2.5 xxxaries, such as countries, regions of countries, and PM10 xxxcities. Ideally, the spatial resolution of the tool and Ozone x x x input data would be matched with the spatial reso- NO xx 2 lution of the assessment context (e.g., using a tool SO2 xx CO x with city-level or finer resolution to assess air pollu- Other Dioxins, Heavy metals, tion impacts in cities). mercury, dioxins, radio Most tools rely upon air quality modeling to es- black nuclides timate exposure, though some may also be able to carbon Health outcome: read in observations from air quality monitors or Mortality xxxdraw information from both models and monitors. (cases) Compared with monitoring, the advantages of us- Disability- xxxing models to simulate air pollutant concentrations adjusted life for health impact assessment include broader spatial years (DALY) or coverage compared to in situ ground-based monitor- years of life ing (though this may not be the case for satellite- lost (YLL) based observations) and the possibility to evaluate Morbidity xxxdifferent future scenarios of emission changes. By (cases) contrast, monitoring data reflect actual ambient lev- els in a specific location for a discrete period in time. Certain tools (e.g., Aphekom, BenMAP-CE) lack of high-quality baseline morbidity rates in many allow users to adjust these monitoring data to re- countries.(22) These types of administrative records flect hypothetical air quality changes (i.e., monitor are generally more challenging to collect than “rollback”). death records. Therefore, while several of the tools Some of the tools reviewed here can be used with global scope have the capability to quantify to estimate air-pollution-related health impacts at morbidity impacts, the capability may be limited to gridded resolution (e.g., BenMAP-CE, EcoSense). certain contexts and applications where high-quality Several of these tools use as an input the results of baseline morbidity rates are available. In addition, full air quality modeling—which in turn accounts extrapolating concentration-response associations for the complex atmospheric chemistry and trans- for morbidity outcomes like hospitalizations and port governing air pollution and also simulates asthma attacks from one population to another is the influence of emission controls on air pollution difficult because health-care access and systems levels. However, application of these tools may be differ widely around the world and both diagnoses prohibitive in some assessment contexts because and coding of diagnoses can be inconsistent. full-scale air quality modeling is generally resource intensive and operating the tool requires signif- icant technical expertise (though web-based and 3.2. Resolution and Exposure Characterization classroom training is often available). The spatial A key difference among the tools reviewed here resolution of the available population data may also is their approach to characterizing population ex- be a limiting factor for fine-scale analyses. Gridded posure to air pollution, changes in exposure result- assessments can typically be aggregated to geopo- ing from emission or concentration changes, and litical boundaries such as cities (though depending whether ambient pollutant data are available in the on the grid resolution, there may only be one or two tool or whether users must specify these data from grid cells for each city), countries, and regions. Survey of Ambient Air Pollution Health Risk Assessment Tools 9

Table V. PM2.5 Concentration-Response, Population, and Baseline Incidence Data Sources for Tools with Regional Scope

Characteristic Aphekom EVA EcoSense

PM2.5 concentration-response relationship: User-defined x American Cancer Society(56) xx Other Several others(57–59) WHO WHO(47) Population data source: User-defined x x Other GEOSTAT JRC population density grid(60) with enhancements Baseline incidence data source: User-defined x World Health Organization x Other EUROSTAT, national statistics

Table VI. Key Operational Characteristics of Tools with livers gridded results while using parameterized air Regional Scope pollution modeling. However, the results may be less able to account for atmospheric chemistry and Characteristic Aphekom EVA EcoSense transport than those based on full-form modeling Format: (i.e., taking the difference between separate model Software download x simulations of a base case and a control case), as they Microsoft office program x typically linearize source-receptor relationships that Web based x Open source x are nonlinear in nature due to nonlinear atmospheric Proprietary x x chemistry. Thus, they may be of limited inter- Peer reviewed/policy applications: pretability in certain contexts (e.g., estimating the Peer reviewed x x x health benefits of reducing SO2 emissions after NOx Used for policy applications x x x emission reductions are in place). While reduced- form tools have some limitations, including in their ability to capture complex atmospheric chemistry When air quality modeling is unavailable, processes, in many cases these limitations can be “reduced-form” tools can generate broad-scale reasonable to accept—for example, when screening a estimates of air pollution impacts. We define here large number of emission control scenarios for which reduced-form tools as those that connect emissions air quality modeling would be resource-prohibitive. to health impacts using built-in parameterizations, Using air quality models for health impact thereby bypassing the need to run expensive and assessment also has several disadvantages, includ- resource-intensive chemical transport modeling. ing that simulated concentrations have inherent Reduced-form tools often rely on built-in relation- uncertainty and that the air quality model may not ships between emissions and the exposure metric have sufficient resolution to match actual exposure (typically concentration) derived from externally patterns (e.g., near-roadway exposures, high urban conducted air quality model simulations. For ex- concentrations). Similarly, modeled concentrations ample, LEAP-IBC relies on influence coefficients may not match the method or spatial resolution of generated by the global chemical transport model, the exposure characterization in the epidemiology GEOS-Chem Adjoint, that links gridded emissions studies from which concentration-response associa- to impacts at the national level. Another exam- tions are drawn, which may introduce error into the ple is the TM5-FASST tool, which is driven by a analysis. Thus, while air quality models are necessary region-to-region source-receptor matrix (i.e., the to address health benefits of alternative future quantified influence of emissions in one region on scenarios across broad spatial scales, simulated con- health impacts in another) that was developed from centrations should be evaluated against observations TM5 global chemical transport model simulations. and care must be taken to match spatial resolutions EcoSense uses country-to-grid matrices derived among the assessment context, air quality model, from various EMEP Unified Model runs, i.e., de- and epidemiological inputs to the health impact 10 Anenberg et al.

Table VII. Source Type and Required User Input (Emissions, Concentration, or Intake Fraction) for Population Exposure Information for Each Tool According to the Categories of Geographical Scope

Exposure Information Source User Input Global Scope Regional Scope

Any concentration input by user Concentration BenMAP-CEa EBD AirQ2.2 IOMLIFET In situ monitor Concentration Aphekomb Global chemical transport model (input by user) Concentration EVA Regional or urban atmospheric chemistry model (input by user) Emissions SIM-Air EVA Reduced-form chemical transport model Emissions LEAP-IBCc EcoSensed TM5-FASSTd Reduced-form econometric model Economic and climate indicators GMAPSe TM f Intake fraction (primary PM2.5 only) Emissions AirCounts aPreloaded with monitor data for the United States and China. bAir quality monitoring data described by Pascal et al.(21) cEmissions are translated to concentrations and impacts using gridded per unit emission influence coefficients. dEmissions are translated to concentrations and impacts using a nationally averaged source-receptor matrix. eInputs are: total primary energy consumption by type of energy, per capita gasoline and diesel consumption, country and city popula- tion, population density, suite of city-specific climate variables, heating degree days, cooling days, gross national income per capita, gross domestic product, technical progress, historical PM, and total suspended particulate (TSP) concentrations where available. f TM (61) The intake fractions used in AirCounts , described by Apte et al., are limited to directly emitted PM2.5 and cannot be used to estimate secondarily formed pollutants (e.g., ozone, secondarily formed PM2.5 components such as sulfate and nitrate). function as closely as possible. Some tools use in situ one of the tools with global scope were preconfig- ground-level monitors, finely resolved population ured with long-term PM2.5 concentration-response information, or remote sensing (e.g., satellite obser- relationships based on findings from the American vations) to improve the performance and resolution Cancer Society Study.(3,56) Three are now using of current concentrations simulated by the model the IERs functions generated for the 2010 GBD (e.g., LEAP-IBC, TM5-FASST). This type of data Study,(14,15) which generate a single concentration- assimilation, however, is not possible for model response curve across the entire range of PM2.5 simulations of future air quality(24,55) or present-day concentrations using ambient air pollution, indoor “what if” scenarios, for which observed data are air pollution, and cigarette smoking studies. By not available. Methods that draw information from contrast, none of the three regional tools use the both monitors and models can improve confidence IERs. Most of the tools do not impose a low- in concentration estimates in such cases. concentration threshold, beyond which health risks per unit pollutant concentration diminish, though many allow a threshold to be specified by the user. 3.3. PM Concentration-Response Relationships, 2.5 Uncertainty in the health impact results may increase Population, and Baseline Incidence at low concentrations (less than approximately 5 Data Sources 3 µg/m PM2.5 annual average) where epidemiolog- As shown in Tables II and V, a relative com- ical data are currently sparse. An assumption of monality among the tools is that many use the linearity at very low concentrations may distort the same sources of data for concentration-response true health impacts of air pollution in relatively relationships, population, and baseline incidence clean atmospheres.(21,24) Contrastingly, applying a rates (e.g., annual number of deaths in a particular low-concentration threshold can understate health population). Several of the global and regional tools impacts at low concentrations if the relationship is are flexible enough to allow users to input data from linear or close to linear. Thus, just as understanding any source for each of these key inputs to the health the shape of the concentration-response curve at high impact function. concentrations is critical for assessments in highly For PM2.5 concentration-response relationships, polluted atmospheres, understanding the shape of three of the global tools allow users to input data the curve at low concentrations may be critical for as- from any source. At the time of this survey, all but sessments in relatively clean atmospheres, such as in Survey of Ambient Air Pollution Health Risk Assessment Tools 11 many economically developed countries. Generally, context. For example, the health benefits of U.S. estimates at such low concentrations impact more air quality policies are generally estimated using on estimates of the burden of current pollution than the most refined tool for the U.S. (BenMAP-CE), on estimates of the benefits of pollution reductions detailed demographic data sets, and the difference (where both old and new concentrations may be between air quality model simulations of a base 3 above 5 µg/m PM2.5 annual average). emissions scenario and a control emission scenario. For population, four of the tools allow users In contrast, it may be time and cost prohibitive to input data from any source, five use data to run air quality modeling for more data- and from Columbia University’s Center for International resource-limited countries; in the absence of more Earth Science Information Network (CIESIN), and refined tools and data sets that are also accessible two use U.N. data for each country. Rather than with limited resources, reduced-form tools (e.g., the these global data sources, the regional tools use data LEAP-IBC) that do not require air quality modeling from any sources or from sources specific to the re- or detailed demographic data sets as inputs may be gion. For baseline incidence rates, four of the global the only way to estimate health benefits of reducing tools allow users to input data from any source, five emissions in those areas. It may be useful to work use World Health Organization (WHO) informa- iteratively, using “lighter” tools initially to scope the tion for regions or countries, and two also use other issues and to see what matters; and moving to more sources of information. Of the regional tools, one resource-intensive tools only if the scoping indicates tool allows users to input baseline incidence infor- that it would be worthwhile to do so. Generally, mation from any source, one is preconfigured with more refined tools and data inputs allow for greater WHO data, and one uses other information sources. certainty and precision in results. Analysts may also consider whether the tool has been peer reviewed, the extent to which analysts 4. KEY OPERATIONAL CHARACTERISTICS have used it to inform policy, and whether it is open The tools also range in format, affecting how ac- source or proprietary. Some tools (e.g., BenMAP, cessible they are to less technical audiences. Some the predecessor to BenMAP-CE) have received tools are client-based software programs, requiring external peer review and have been “exercised” users to download and install the software to use extensively in the course of supporting national air it. These tools include extensive data sets of health quality regulations (e.g., U.S. EPA National Ambi- impact functions, population, and health data that ent Air Quality Standards). The majority of the tools users may modify, but are generally more compli- are either currently open source or open-source cated to use and may require users to invest time versions are in development. A critical advantage and resources in training themselves. Other tools run of open-source tools over those that are proprietary within Microsoft Excel, which is generally accessi- is that they are fully transparent, allowing analysts ble to most users but may require them to purchase to evaluate the underlying algorithms and data sets the Microsoft Office suite. Since many analysts are used to calculate impacts. familiar with Microsoft Office, these tools may not Finally, analysts may consider whether the tool is require extensive training materials. A few tools are maintained as a “living” tool, or whether the included web based, enabling users to generate air pollution data sets and methods are fixed in place or obsoles- health impact estimates without downloading or in- cent. The data inputs to air pollution health impact stalling a program. Web-based tools may be most ac- assessments are often updated over time to reflect cessible to nontechnical users, particularly in coun- changes in the science. For example, the size of the tries that lack the resources to conduct full-scale, de- population exposed to air pollution is a major driver tailed, and very refined health impact assessments. of air pollution health impacts, and updating data sets Users may also wish to consider the complexity over time can capture growth, aging, and migration of the tool (e.g., data inputs and resources required) changes. Similarly, updating baseline mortality and and the extent to which it is preloaded with the data disease rates over time can capture the “epidemio- needed to perform an assessment. The range of tools logical transition” from infectious disease to chronic described here reflects a range of technical com- disease as economies develop. Air pollution expo- plexity and accessibility. Users will want to consider sure levels are also changing rapidly as economies balancing their tolerance for technical complexity develop and urbanization occurs, and exposure char- with the level of specificity called for in the policy acterization methods can be updated to reflect the 12 Anenberg et al. latest emissions, meteorology, and atmospheric methods of each. However, additional uncertainties chemistry information. On a more operational level, exist as to the shape of the concentration-response some tools requiring software downloads (e.g., curve at different concentration levels (i.e., the slope AirQ2.2) may not function on updated operating of the concentration-response factor at different systems. concentrations), the extent to which different air pol- lutant mixtures pose more or less risk, and the degree to which concentration-response relationships found 5. CHALLENGES in one population can be extrapolated to others with Despite significant advancements in quantifying different lifestyles, age structures, and medical care the health impacts of ambient air pollution over the (e.g., from a U.S. cohort to other countries). last decade, several uncertainties and information In addition to uncertainties in the concentration- gaps remain. This section describes key uncertainties response association, exposure estimates are subject that propagate throughout the air pollution health to uncertainties in the magnitude and spatial dis- impact assessment methodology (Section 5.1), the tribution of emissions, chemical and physical pro- degree to which ambient air pollution health impact cesses influencing the impact of emissions on pol- assessment tools have been integrated with other lutant concentrations, and the spatial (horizontal) health risks (Section 5.2), and challenges in interpret- and altitudinal (vertical) resolution. However, uncer- ing technical information generated by air pollution tainty characteristics of many air quality models are health impact assessments for use in policy decisions difficult to ascertain. For assessments of future health (Section 5.3). impacts, uncertainties in socioeconomic assumptions such as economic growth and population health are also important. Although we are more confident in 5.1. Uncertainty and Information Gaps recent estimates of population size and spatial dis- Air pollution health impact assessment combines tribution, other demographic parameters, including information from different sources, including esti- the baseline mortality and morbidity rates around mated pollutant exposure, demographics, and the the world, are uncertain. Projecting future demo- relationship between ambient concentrations and graphic changes is subject to uncertainties regarding health outcomes. Each of these information sources population aging, migration, and the epidemiological carries with it some degree of uncertainty that in- transition from infectious disease to chronic diseases, fluences the precision and confidence in the health which are more affected by air pollution exposure. impact results (Fig. 2). These uncertainties propa- Because their magnitude is unknown and not gate as the health impact assessment moves through estimated, the uncertainty in each of these param- each stage, with the relative influence of each param- eters is often excluded from the characterization of eter’s uncertainty dependent on the concentration- uncertainty in the health impact assessment results, response association used to quantify health impacts which may give a misimpression as to the precision as well as the assessment context. Characterizing the and confidence of the results. This is particularly im- total uncertainty in the health risk and impact results, portant in assessments for which some unquantified and representing it in ways that are both accessible sources may be orders of magnitude more impor- and accurate, is challenging because the magnitude tant to the results than the standard error in the of uncertainty in each of these individual information concentration-response association. For example, sources is often unknown and not estimated. global health impact assessments that incorporate Many air pollution health impact assessments the standard error from the concentration-response express the quantitative level of uncertainty by calcu- association into the confidence intervals around lating a confidence interval using the standard error the results exclude a likely much larger source of from the epidemiologically-derived concentration- uncertainty from inaccurate global air pollution response relationship. Since the differences between emissions and concentrations, as well as the extrap- epidemiological effect coefficients are typically olation of concentration-response associations from larger than the uncertainty of individual effect one country to the rest of the world.(22–24) The 2010 coefficients,(62) some tools (e.g., BenMAP-CE) allow Global Burden of Disease analysis represented an for simple pooling techniques (e.g., fixed effect and improvement in characterizing uncertainties from random effects) to combine multiple effect coeffi- multiple sources as it reported confidence intervals cients, leveraging the different populations and study that reflected uncertainty from two input parameters: Survey of Ambient Air Pollution Health Risk Assessment Tools 13

Fig. 2. Sources of uncertainty affecting quantifi- cation of air-pollution-related health impacts. RR = relative risk.

(1) the predicted air quality levels and (2) the ef- by the air quality model.(17) Sensitivity analyses can fect coefficient from the epidemiological study.(15) also provide useful information as to the influence of However, the study did not quantitatively address error in various parameters.(24) Sensitivity analysis is uncertainty in other parameters, such as population particularly useful for parameters for which observa- mortality rates. Contrasting global-scale assessments tions to compare against do not exist (e.g., projected with U.S. assessments where information is rich and future demographics). Tools that allow for users to air quality monitors extensive, the parameter that alter individual parameters can be used effectively to has the largest influence on health impact results has conduct sensitivity analyses; those that more rigidly been found to be the effect coefficient, followed by build in data sets that cannot be altered are less able the air quality change examined.(62) These two ex- to be used for this purpose. Future assessments that amples of global and U.S. assessments demonstrate incorporate multiple sources of uncertainty into the that the most influential uncertainty sources likely results can provide a more complete indication of differ between assessment contexts. the uncertainty magnitude. Regardless of the extent None of the tools surveyed here are capable to which uncertainties are quantified, it is useful of fully accounting for all sources of uncertainty, to discuss all uncertainty sources qualitatively with though the limitation lies mainly with the lack of some interpretation by the analyst as to the potential information about input parameter uncertainty as importance of each uncertainty source. opposed to building into the tools the ability to statistically combine multiple uncertainty sources. In the absence of quantitative uncertainty estimates 5.2. Integrating Ambient Air Pollution Health in each parameter, analysts may consider other Impacts with Other Health Risks methods of addressing error in the input parameters. Comparative risk assessments that account For example, benchmarking simulated air pollution for multiple health stressors, such as indoor air concentrations against readings from in situ, satellite, pollution, cigarette smoking, climate change, vehicle and other observatory techniques and adjusting the accidents, and physical activity, can provide greater concentration levels used as input to the health im- understanding and context for the impacts of mul- pact assessment can help minimize bias introduced tiple risk factors on society.(15) The tools reviewed 14 Anenberg et al. here focus on ambient air pollution, but several Other tools can assess a variety of health risks new tools seek to quantify household air pollution within the same framework, but have not included health impacts and capture the interplay between air the capability to assess air pollution health impacts. pollution and other health risks. Several tools under The Lives Saved Tool (LiST; http://www.jhsph.edu/ development compile information on household departments/international-health/centers-and-institu air pollution exposures in different countries (e.g., tes/institute-for-international-programs/list/index. World Health Organization global database of html), which allows users to estimate the global household air pollution measurements, available health benefits of public health interventions (e.g., at http://www.who.int/indoorair/health_impacts/ vitamin A supplementation and malaria treatments), databases_iap/en/). Other tools are being developed may soon include the ability to estimate the benefits to go beyond household air pollution exposure of household air pollution interventions.(63) The characterization to health impacts. For example, the Health Economic Assessment Tool (http://www.heat Household Air Pollution Impacts Tool (HAPIT, walkingcycling.org) includes the impact of increased available at https://hapit.shinyapps.io/HAPIT/) de- walking and cycling on health, but does not currently veloped by University of California–Berkeley allows include the capability to quantify air pollution users to input reductions in indoor air pollution health impacts. Expand the existing tools to quantify exposure concentrations to estimate premature air pollution and other health risks in a rigorous deaths and DALYs avoided in every country. The and integrated way can provide more complete tool also includes the ability to compare benefits information to decisionmakers and others. against mitigation costs to generate estimates of As air pollution and climate are interrelated in cost effectiveness. Since household emissions can several ways, tools may also increasingly consider contribute significantly to outdoor air pollution in both of these health stressors together. In addition to many places around the world, this type of tool air pollutants that contribute to climate change (e.g., could be integrated with ambient air pollution health black carbon, ozone) and the influence of climate impact tools to assess the total benefits of household change on air pollution (e.g., via changing emissions air pollution mitigation due to exposure both indoors and meteorology), health-harmful air pollutants and and outdoors, avoiding double counting. long-lived greenhouse gases like CO2 are often emit- Policies affecting air quality can also influence ted by the same sources. Therefore, some mitiga- other sources of risk. For example, encouraging com- tion measures will likely influence both air pollu- muters to substitute bicycles in place of automobiles tion and climate change simultaneously.(24,26,31) To may reduce air pollution and bring health benefits account for the multiple impacts of changing emis- due to physical exercise but increase risk of traffic sions, the LEAP-IBC quantifies impacts of emission accidents. To date, only a couple of tools are ca- changes on human health, climate change, and agri- pable of assessing air pollution health impacts and cultural yields simultaneously. Some evidence also other population health factors (including cigarette suggests that the health effects of air pollutants are smoking, vehicle accidents, noise, and physical ac- modified by temperature, indicating that the health tivity) within the same framework. The Interna- risks imposed by climate change and air pollution tional Futures project at the University of Denver may be synergistic.(64,65) Tools that integrate climate (http://pardee.du.edu/) integrates household air pol- change and air pollution impacts therefore provide a lution and ambient air pollution impacts into a much more comprehensive accounting of impacts that can broader global model that includes economic, health, be informative for policy decisions.(55) environmental, technological, and other changes over time. The IOMLIFET model can incorporate 5.3. Interpreting Technical Information for relative risks for any risk factor, including air pollu- Policy Decisions tion and other health risks—provided that the end user can provide the appropriate risk parameters. Another challenge not fully accounted for in Finally, the Integrated Transport and Health Im- these tools is interpreting and reporting the results pact Modeling Tool (ITHIM) for the United King- such that they are well matched to the objective of dom integrates health impact assessment of transport the risk assessment. The types of results provided through changes in physical activity, road traffic in- and how they are explained may differ for different jury risk, and urban air pollution (see Supplemental assessment contexts. For example, for assessments Material). intended to provide information about the burden Survey of Ambient Air Pollution Health Risk Assessment Tools 15

Table VIII. Example Assessment Contexts, Key Technical and Operational Considerations When Selecting an Air Pollution Health Impact Assessment Tool, and Potentially Appropriate Tools to Use for the Assessment

Assessment Context Key Considerations Potentially Appropriate Tools

National ambient air quality standards Maximize technical rigor. Resources are AirQ2.2 regulatory support in data-rich country typically available for air quality modeling Aphekom (Europe) or measurements. Standards are BenMAP-CE concentration based rather than emissions EBD (Europe) based. EVA (Northern Hemisphere) IOMLIFET National-scale tools not included in this survey National ambient air quality standards Resources for air quality modeling or AirQ2.2 regulatory support in data-poor country measurements are typically not available. Aphekom (Europe) Reduced-form tools can provide an order BenMAP-CE global module of magnitude assessment. Standards are concentration based rather than emissions based.

Hypothetical reduction of PM2.5 This hypothetical scenario does not require AirQ2.2 concentrations to WHO guidelines information about the magnitude of Aphekom (Europe) emission reductions. The tool may input BenMAP-CE global module (global) concentrations rather than emissions. EVA Reduced-form tools can provide an order of magnitude assessment. Reduced air pollution emissions resulting Resources may be available for global AirQ2.2 from multi-governmental agreements chemical transport modeling, but BenMAP-CE (e.g., on climate change, long-range reduced-form tools can provide an order IOMLIFET transboundary air pollution) of magnitude assessment. Analysis starts LEAP-IBC with emissions rather than concentrations. TM5-FASST GMAPS EVA TM Reduced air pollution emissions from Project developers typically responsible for AirCounts (primary PM2.5 in urban international development projects (e.g., providing information about the benefits areas) World Bank, Global Environment of the project in their application are Aphekom (Europe) Facility) unlikely to have air quality modeling BenMAP-CE global module expertise. Reduced-form tools can provide LEAP-IBC an order of magnitude assessment. Spatial EcoSense (Europe) resolution must most closely match the SIM-Air scale of the project (typically urban). EVA

of disease due to air pollution exposure for non- formed decisions if they have an appreciation of what regulatory purposes, it may be sufficient to provide information is and is not represented by the values only a limited set of quantitative estimates, includ- provided to them. ing cases of death and disease.(15,17) Conversely, as- The tools surveyed here produce results in a va- sessments intended to inform the setting and design riety of formats, including different types of numer- of regulations may require more extensive analyses ical results, descriptions, charts, graphs, and maps. and results, including the percentage reduction in the They range in the extent of explanations given for air pollution health burden, various sensitivity anal- each result type, and the extent to which they rely on yses, uncertainty estimation, and all assumptions un- the analyst to understand the parameters, equations, derlying these results.(45) Risk analysts can more ef- and results on their own. The challenge is perhaps fectively communicate if they thoroughly understand greatest with reduced-form tools that have built- the data sets, calculations, and assumptions used to in parameters that the user cannot change; these produce risk results, as well as which values and anal- parameters are typically produced using an external yses to highlight.(66) Policymakers can make more in- model during the development of the tool and users 16 Anenberg et al. of these tools may not be aware of or understand how tools demonstrates that there is an important trade- they were produced. While easily applied tools are off between technical refinement and accessibility more accessible, if not fully understood by the end for a broad range of applications. For some purposes, user, they may enable misuse or misinterpretation of it may be sufficient to use a coarsely resolved global findings, potentially leading to poorly designed poli- tool based on reduced-form air quality modeling cies. Overly automated tools thus have the potential given data and resource and time limitations. Even to undermine the ability of analysts to fully under- in geographical areas for which high-quality data stand their own assessments. Tools that are flexible exist, reduced-form tools may be useful to screen enough to allow users to input their own data sets, many potential emission scenarios, identifying those tools that are open source, and tools that are exten- that may benefit from more detailed evaluation. sively documented and peer reviewed can ease this However, where possible, more finely resolved and challenge significantly. In the future, guidance could sophisticated tools based on full air quality modeling assist users in interpreting numeric results from air allow for greater confidence and precision in results, pollution health impact assessment tools and com- particularly useful for regulatory analysis. And when municating results to decisionmakers. Where possi- sufficient air pollution measurements, provided by ble, these tools could be used in the context of train- air quality monitoring networks, are available, they ing of environmental health professionals to be able also provide a reliable source of a population’s to share the philosophy behind the tool, and this type exposure.(20) Given the heterogeneity among tech- of guidance could be incorporated into training ma- nical and operational characteristics, analysts may terials for individual tools with specific application to wish to select tools that provide the appropriate the type of results (e.g., avoided mortality cases, life geographic scope, resolution, and maximum degree expectancy changes, percent reduction in air pollu- of technical rigor within the resource (e.g., data, tion mortality burden) produced by the tool. time, technical capability) constraints. Matching the abilities of individual tools with specific assessment contexts could highlight ways in 6. CONCLUSION which the currently available tools could be im- This article reviews 12 current and publicly avail- proved or whether new methods or tools are needed able multinational tools that combine air quality in- to fill a specific need. To demonstrate the techni- formation, epidemiologically-derived concentration- cal and operational considerations when selecting response associations, and demographic data sets a tool to perform the assessment, Table VIII pro- to estimate air-pollution-related health risks. Nine vides an initial matching of potentially appropriate of the tools are capable of assessing air pollution tools to example assessment contexts. This table does health risks in cities, countries, and/or regions around not represent the variety of assessment contexts that the world or on a gridded resolution at any ge- these tools are often asked to inform, nor does it ographical scope from local to global (we define reflect consensus among the tool developers as to these to be global in scope). Three of the tools en- how their tools are most appropriately used. Nev- compass several countries (we define these to be ertheless, it demonstrates how specific tools can be regional in scope). The tools share several com- matched to different assessment contexts. A system- mon attributes. Nearly all estimate PM2.5 impacts atic intercomparison among the inputs, assumptions, (though two include only the directly emitted com- calculations, and results of the various health risk ponents of PM2.5), consider mortality outcomes, and assessment tools surveyed here would help analysts are open source. Many of the tools also use similar identify the tool most appropriate for different as- data sources for concentration-response associations, sessment contexts. population, and baseline mortality rates. The tools In addition to a systematic tool intercomparison, also vary in important respects, including the expo- there are opportunities to better account for mul- sure information sources, format, and technical com- tiple sources of uncertainty and to integrate health plexity. risks from multiple stressors, including ambient out- Different tools are appropriate for different door air, indoor air pollution, cigarette smoking, cli- assessment contexts, and analysts must consider mate change, vehicle accidents, and physical activity. the technical and operational specifications of the Such integrated risk assessments can provide greater tool necessary to meet the needs of the assessment understanding and context for the impacts of multi- context. The range of key characteristics among the ple risk factors on society. In addition, increasing the Survey of Ambient Air Pollution Health Risk Assessment Tools 17 extent to which these tools and their underlying data, 14/006. Washington, DC: U.S. Environmental Protection formulas, and documentation are publicly available Agency, 2015. 10. Health Effects Institute (HEI). Special Report 17: Traffic- and open source can enhance transparency, repro- Related Air Pollution: A Critical Review of the Literature on ducibility, and the ability for outside investigators to Emissions, Exposure and Health Effects. Boston, MA: Health make improvements. Effects Institute, 2010. 11. Laumbauch RJ, Kipen HM. Respiratory health effects of air pollution: Update on biomass smoke and traffic pollu- ACKNOWLEDGMENTS tion. Journal of Allergy Clinical Immunology, 2012; 129(1): 3–11. The authors alone are responsible for the views 12. Delfino RJ, Wu J, Tjoa T, Gullesserian S, Nickerson B, Gillen DL. Asthma morbidity and ambient air pollution: Effect mod- expressed in this publication and they do not nec- ification by residential traffic-related air pollution. Epidemiol- essarily represent the decisions or stated policies ogy, 2014; 25(1):48–57. of their employers. This article was expanded from 13. Wesson K, Fann N, Morris M, Fox T, Hubbell B. A multi- pollutant risk-based approach to air quality management: a white paper prepared while S. Anenberg was Case study for Detroit. Atmospheric Pollution Research, employed by the U.S. Environmental Protection 2010; 1:296–304. Agency. We thank Ed Hanna, Daven Henze, Denise 14. Fann N, Wesson K, Hubbell B. Characterizing the confluence of air pollution risks in the United States. Air Quality Atmo- Mulholland, Nicholas Muller, Dave Stieb, Marko sphere, and Health, 2015; doi: 1007/s11869-015-0340-9. Tainio, Harry Vallack, and Jason West for useful 15. Burnett RT, Pope CA, Ezzati M, Olives C, Lim SS, Mehta contributions. We appreciate helpful comments re- S, Shin HH, Singh G, Hubbell B, Brauer M, Anderson HR, Smith KR, Balmes JR, Bruce NG, Kan H, Laden F, Pruss- ceived during the WHO Expert Meeting on Health Ustun A, Turner MC, Gapstur SM, Diver WR, Cohen A. Risk Assessment and Convention on Long-Range An integrated risk function for estimating the global burden Transboundary Air Pollution Task Force on Health of disease attributable to ambient fine particulate matter ex- posure. Environmental Health Perspective, 2014; 122(4):397– meetings in Bonn, Germany, May 12–14, 2014. 403. 16. Lim SS, Vos T, Flaxman AD, Danaei G, Shibuya K, Adair- Rohani H, et al. A comparative risk assessment of burden of REFERENCES disease and injury attributable to 67 risk factors and risk fac- tor clusters in 21 regions, 1990–2010: A systematic analysis 1. Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F. for the Global Burden of Disease Study 2010. Lancet, 2013; Ozone and short-term mortality in 95 US urban communities, 380(9859):2224–2260. 1987–2000. JAMA, 2004;292(19):2372–2378. 17. Hubbell BJ, Hallberg A, McCubbin D, Post E. Health-related 2. Jerrett M, Burnett RT, Pope CA III, Ito K, Thurston G, benefits of attaining the 8-hr ozone standard. Environmental Krewski D, Shi Y, Calle E, Thun M. Long-term ozone expo- Health Perspective, 2005; 113(1):73–82. sure and mortality. New England Journal of Medicine, 2009; 18. Fann N, Lamson AD, Anenberg SC, Wesson K, Risley D, 360:1085–1095. Hubbell BJ. Estimating the national public health burden as- 3. Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, sociated with PM2.5 and ozone. Risk Analysis. 2011; 32(1):81– Turner MC, Pope CA III, Thurston G, Calle EE, Thun MJ, 95. Beckerman B, DeLuca P, Finkelstein N, Ito K, Moore DK, 19. Anderson HR, Ruggles R, Pandey KD, Kapetanakis V, Newbold KB, Ramsay T, Ross Z, Shin H, Tempalsi B. Ex- Brunekreef B, Lai CK, Strachan DP, Weiland SK. Ambi- tended Follow-Up and Spatial Analysis of the American Can- ent particulate pollution and the world-wide prevalence of cer Society Study Linking Particulate Air Pollution and Mor- asthma, rhinoconjunctivitis and eczema in children: Phase one tality. Boston: Health Effects Institute, 2009. of the International Study of Asthma and Allergies in Child- 4. Lepeule J, Laden F, Dockery D, Schwartz J. Chronic exposure hood (ISAAC). Occupational and Environmental Medicine, to fine particles and mortality: An extended follow-up of the 2009; 67(5):293–300. Harvard Six Cities Study from 1974 to 2009. Environmental 20. Henschel S, Atkinson R, Zeka A, Le Tertre A, Analitis A, Health Perspectives, 2012; 120(7):965–970. Katsouyanni K, Chanel O, Pascal M, Forsberg B, Medina S, 5. World Health Organization. Air Quality Guidelines. Global Goodman PG. Air pollution interventions and their impact Update 2005. Copenhagen, Demark: World Health Organiza- on public health. International Journal of Public Health, 2012; tion, 2006. 57(5):757–768. 6. U.S. Environmental Protection Agency (U.S. EPA). Inte- 21. Pascal M, Corso M, Chanel O, Declercq C, Badaloni C, Cesa- grated Science Assessment (ISA) for Sulfur Oxides — Health roni G, Henschel S, Meister K, Haluza D, Martin-Olmedo P, Criteria (Final Report). EPA/600/R-08/047F. Washington, Medina S. Assessing the public health impacts of urban air pol- DC: U.S. Environmental Protection Agency, 2008. lution in 25 European cities: Results of the Aphekom project. 7. U.S. Environmental Protection Agency (U.S. EPA). Inte- Science of Total Environment, 2013; 449:390–400. grated Science Assessment for Particulate Matter (Final Re- 22. Cohen AJ, Anderson HR, Ostro B, Pandey KD, port). EPA/600/R-08/139F. Washington, DC: U.S. Environ- Krzyzanowski M, Kunzli N, Gutschmidt K, Pope CA mental Protection Agency, 2009. III, Romieu I, Samet JM, Smith KR. Urban air pollution. Pp. 8. U.S. Environmental Protection Agency (U.S. EPA). Inte- 1353–1434 in Ezzati M, Lopez AD, Rodgers A, Murray CJL grated Science Assessment of Ozone and Related Photochem- (eds). Comparative Quantification of Health Risks: Global ical Oxidants (Final Report). EPA/600/R-10/076F. Washing- and Regional Burden of Disease Due to Selected Major Risk ton, DC: U.S. Environmental Protection Agency, 2013. Factors. Geneva: World Health Organization, 2004. 9. U.S. Environmental Protection Agency (U.S. EPA). Inte- 23. Anenberg SC, Horowitz LW, Tong DQ, West JJ. An esti- grated Science Assessment for Oxides of Nitrogen (Health mate of the global burden of anthropogenic ozone and fine Criteria) — Second External Review Draft. EPA/600/R- particulate matter on premature human mortality using atmo- 18 Anenberg et al.

spheric modeling. Environmental Health Perspectives, 2010; 36. Fann N, Nolte CG, Dolwick P, Spero TL, Curry Brown A, 118(9):1189–1195. Phillips S, Anenberg S. The geographic distribution and eco- 24. Anenberg SC, Talgo K, Arunachalam S, Dolwick P, Jang C, nomic value of climate change-related ozone health impacts West JJ. Impacts of global, regional, and sectoral black car- in the United States in 2030. Journal of the Air & Waste Man- bon emission reductions on surface air quality and human agement Association, 2015; 65(5):570–580. mortality. Atmospheric Chemistry and Physics, 2011; 11:7253– 37. Nawadha A. Reductions of PM2.5 air concentrations and pos- 7267. sible effects on premature mortality in Japan. Water, Air, & 25. Anenberg SC, Schwartz J, Shindell D, Amann M, Faluvegi G, Soil Pollution, 2013; 224:1508. Klimont Z, Janssens-Maenhout G, Pozzoli L, Van Dingenen 38. Guttikunda SK, Jawahar P. Application of SIM-air modeling R, Vignati E, Emberson L, Muller NZ, West JJ, Williams tools to assess air quality in Indian cities. Atmospheric Envi- M, Demkine V, Hicks WK, Kuylenstierna J, Raes F, Ra- ronment, 2012; 62:551–561. manathan V. Global air quality and health co-benefits of mit- 39. Brandt J, Silver JD, Christensen JH, Andersen MS, Bønløkke igating near-term climate change through methane and black J, Sigsgaard T, Geels C, Gross A, Hansen AB, Hansen KM, carbon emission controls. Environmental Health Perspectives, Hedegaard GB, Kaas E, Frohn LM. Contribution from the 2012; 120(6):831–839. ten major emission sectors in Europe to the health-cost ex- 26. Shindell D, Faluvegi G, Walsh M, Anenberg SC, Van Din- ternalities of air pollution using the EVA model system — An genen R, Muller NZ, Austin J, Koch D, Milly G. Climate, integrated modelling approach. Atmospheric Chemistry and health, agricultural, and economic impacts of tighter vehicle- Physics, 2013; 13:7725–7746. emission standards. Nature Climate Change, 2011; 1:59–66. 40. Guttikunda SK, Goel R. Health impacts of particulate pollu- 27. West JJ, Smith SJ, Silva RA, Naik V, Zhang Y, Adelman tion in a megacity — Delhi, India. Environmental Develop- Z, Fry MM, Anenberg S, Horowitz LW, Lamarque J-F. Co- ment, 2013; 6:8–20. benefits of mitigating global greenhouse gas emissions for fu- 41. Guttikunda SK, Khaliquzzaman M. Health benefits of adapt- ture air quality and human health. Nature Climate Change, ing cleaner brick manufacturing technologies in Dhaka, 2013; 3:885–889. Bangladesh. Air Quality Atmosphere and Health, 2014; 28. Chambliss SE, Silva R, West JJ, Zeinali M, Minjares R. Es- 7(1):103–112. timating source-attributable health impacts of ambient fine 42. Guttikunda SK, Lodoysamba S, Bulgansaikhan B, Dashdon- particulate matter exposure: Global premature mortality from dog B. Particulate pollution in Ulaanbaatar, Mongolia. Air surface transportation emissions in 2005. Environmental Re- Quality Atmosphere and Health, 2013; 6(3):589–601. search Letters, 2014; 9:104009. 43. Kheirbeck I, Wheeler K, Walters S, Kass D, Matte T. PM2.5 29. Anenberg SC, West JJ, Fiore AM, Jaffe DA, Prather MJ, and ozone health impacts and disparities in New York City: Bergmann D, Cuvelier K, Dentener FJ, Duncan BN, Gauss Sensitivity to spatial and temporal resolution. Air Quality At- M, Hess P, Jonson JE, Lupu A, MacKenzie IA, Marmer E, mosphere and Health, 2013; 6(2):473–486. Park RJ, Sanderson MG, Schultz M, Shindell DT, Szopa S, 44. Holland M, Hunt A, Hurley F, Navrud S, Watkiss P. Service Vivanco MG, Wild O, Zeng G. Intercontinental impacts of contract for carrying out cost-benefit analysis of air quality ozone pollution on human mortality. Environmental Science related issues, in particular in the Clean Air for Europe & Technology, 2009; 43:6482–6487. (CAFE) Programme. Methodology for the cost-benefit anal- 30. Anenberg SC, West JJ, Yu H, Chin M, Schulz M, Bergmann ysis for CAFE: Volume 1: Overview of methodology. Web D,BeyI,BianH,DiehlT,FioreA,HessP,MarmerE,Mon- published: AEA Technology Environment, 2005. Available at: tanaro V, Park R, Shindell D, Takemura T, Dentener F. Im- http://europa.eu.int/comm/environment/air/cafe/activities/cba. pacts of intercontinental transport of anthropogenic fine par- htm. ticulate matter on human mortality. Air Quality, Atmosphere 45. Amann M (ed). Policy scenarios for the revision of the The- and Health, 2014; 7(3):369–379. matic Strategy on Air Pollution. TSAP Report #10. Laxen- 31. Brandt J, Silver JD, Christensen JH, Andersen MS, Bønløkke burg, Austria: International Institute for Applied Systems J, Sigsgaard T, Geels C, Gross A, Hansen AB, Hansen KM, Analysis, March 2013. Hedegaard GB, Kaas E, Frohn LM. Assessment of past, 46. U.S. Environmental Protection Agency (U.S. EPA). Regula- present and future health-cost externalities of air pollution tory Impact Analysis for the Final Revisions to the National in Europe and the contribution from international ship traf- Ambient Air Quality Standards for Particulate Matter. EPA- fic using the EVA model system. Atmospheric Chemistry and 452/R-12-005. Washington, DC: U.S. Environmental Protec- Physics. 2013b; 13:7747–7764. tion Agency, 2012. 32. Likhvar VN, Pascal M, Markakis K, Colette A, Hauglaustaine 47. World Health Organization. Health Risks of Air Pollu- D. A multi-scale health impact assessment of air pollution tion in Europe — HRAPIE Project. Recommendations for over the 21st century. Science of the Total Environment, 2015; Concentration–Response Functions for Cost–Benefit Analysis 514:439–449. of Particulate Matter, Ozone and Nitrogen Dioxide. Copen- 33. He K, Lei Y, Pan X, Zhang Y, Zhang Q, Chen D. Co-benefits hagen, Denmark: World Health Organization, 2013. from energy policies in China. Energy, 2010; 35(11):4265– 48. Le Tertre A, Schwartz J, Touloumi G. Empirical Bayes and 4272. adjusted estimates approach to estimating the relation of mor- 34. Boldo E, Linares C, Lumbreras J, Borge R, Narros A, Garcia- tality to exposure of PM. Risk Analysis, 2005; 25(3):711–718. Perez J, Fernandez-Navarro P, Perez-Gomez B, Aragones N, 49. Fiore A, Dentener FJ, Wild O, Cuvelier C, Schultz MG, et al. Ramis R, Pollan M, Moreno T, Karanasiou A, Lopez-Abente Multimodel estimates of intercontinental source-receptor re- G. Health impact assessment of a reduction in ambient PM2.5 lationships for ozone pollution. Journal of Geophysical Re- levels in Spain. Environment International, 2011; 37(2):342– search, 2009; 114:D04301. 348. 50. Rao ST, Galmarini S, Puckett K. Air quality model evaluation 35. Boldo E, Linares C, Aragones N, Lumbreras J, Borge R, de la international initiative (AQMEII). Bulletin of the American Paz D, Perez-Gomez B, Fernandez-Navarro P, Garcia-Perez J, Meteorological Society, 2011;92:23–30. Pollan M, Ramis R, Moreno T, Karanasiou A, Lopez-Abente 51. Lamarque J-F, Shindell DT, Josse B, Young PJ, Cionni I, G. Air quality modeling and mortality impact of fine partic- Eyring V, Bergmann D, Cameron-Smith P, Collins WJ, Do- ulate matter reduction policies in Spain. Environmental Re- herty R, Dalsoren S, Faluvegi G, Folberth G, Ghan SJ, search, 2014; 128:15–26. Horowitz LW, Lee YH, MacKenzie IA, Nagashima T, Naik Survey of Ambient Air Pollution Health Risk Assessment Tools 19

V, Plummer D, Righi M, Rumbold S, Schulz M, Skeie RB, 60. Gallego FJ. A population density grid of the European Union. Stevenson DS, Strode S, Sudo K, Szopa S, Voulgarakis A, Population and Environment, 2010; 31:460–473. Zeng G. The Atmospheric Chemistry and Climate Model In- 61. Apte J, Emilie B, Julian M, William N. Global intraurban tercomparison Project (ACCMIP): Overview and description intake fractions for primary air pollutants from vehicles and of models, simulations and climate diagnostics. Geoscientific other distributed sources. Environmental Science & Technol- Model Development, 2013; 6:179–206. ogy, 2012; 46(6):3415–3423. 52. World Health Organization. WHO Expert Meeting: Methods 62. Mansfield C, Sinha P, Henrion M. Influence analysis in and Tools for Assessing the Health Risks of Air Pollution support of characterizing uncertainty in human health at Local, National, and International Level. Meeting Report. benefits analysis, Final Report. 2009; U.S. Environ- Copenhagen, Denmark: World Health Organization, 2014. mental Protection Agency. Available at: http://www.epa. 53. Pope CA III, Cropper M, Coggins J, Cohen A. Health bene- gov/ttn/ecas/regdata/Benefits/influence_analysis_final_report_ fits of air pollution abatement policy: Role of the shape of the psg.pdf, Accessed April 17, 2015. concentration-response functions. Journal of the Air & Waste 63. Bruce NG, Dherani MK, Das JK, Balakrishnan K, Adair- Management Association, 2015; 65(5):516–522. Rohani H, Bhutta ZA, Pope D. Control of household air pol- 54. Miller BG, Hurley JF. Life table methods for quantitative im- lution for child survival: Estimates for intervention impacts. pact assessments in chronic mortality. Journal of Epidemiol- BMC Public Health, 2013; 13(Suppl 3):S8. ogy & Community Health, 2003; 57:200–206. 64. Ren C, Williams G, Mengersen K. Temperature enhanced ef- 55. Geels C, Andersson C, Hanninen¨ O, Lansø AS, Schwarze fects of ozone on cardiovascular mortality in 95 large US com- P, Brandt J. Future premature mortality due to air pollu- munities, 1987–2000: Assessment using the NMMAPS data. tion in Europe — Sensitivity to changes in climate, anthro- Archives of Environmental & Occupational Health, 2009; pogenic emissions, population and building stock. Interna- 64(3):177–184. tional Journal of Environmental Research and Public Health, 65. Jhun I, Fann N, Zanobetti A, Hubbell B. Effect modification 2015; 12:2837–2869. of ozone-related mortality by temperature in 97 US cities. En- 56. Pope CA, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito vironment International, 214; 73:128–134. K, Thurston GD. Lung cancer, cardiopulmonary mortality and 66. Medina S, Le Tertre A, Saklad M. On behalf of the Apheis long-term exposure to fine particulate air pollution. JAMA, collaborative network. The Apheis project: Air pollution and 2002; 287:1132–1141. health — A European information system. Air Quality Atmo- 57. Anderson HR, Atkinson WA, Peacock JL, Marston L, Kon- sphere and Health, 2009; 2(4):185–198. stantinou K. Meta-Analysis of Time-Series Studies and Panel Studies of Particulate Matter (PM) and Ozone (O3). Report of a WHO Task Group. Copenhagen: World Health Organi- SUPPORTING INFORMATION zation, 2004. 58. Atkinson RW, Anderson HR, Sunyer J, Ayres J, Baccini M, Additional supporting information may be found in Vonk JM, Boumghar A, Forastiere F, Forsberg B, Touloumi G, Schwartz J, Katsouyanni K. Acute effects of particulate air the online version of this article at the publisher’s pollution on respiratory admissions. American Journal of Res- website: piratory and Critical Care Medicine, 2005; 164(10):1860–1866. 59. Gryparis A, Forsberg B, Katsouyanni K, Analitis A, Touloumi G, Schwartz J, Samoli E, Medina S, Anderson HR, Niciu Online Supplement EM, Wichmann HE, Kriz B, Kosnik M, Skorkovski J, Vonk Table S1. Key Technical Characteristics of Tools JM, Dortbudak Z. Acute effects of ozone on mortality from with National Scope the “Air pollution and health a European approach” Project. American Journal of Respiratory Critical Care Medicine, Table S2. Key Operational Characteristics of Tools 2004; 170(10):1080–1087. with National Scope